Atmospheric pollution source tracing method and system fusing numerical model and artificial intelligence algorithm

Abstract

The application discloses an atmospheric pollution tracing method fusing a numerical model and an artificial intelligence algorithm, and relates to the technical field of atmospheric pollution tracing. The method preprocesses meteorological data and simulates a high-precision meteorological field, converts a result into a readable format of a Lagrangian diffusion model, generates a high-resolution space-time footprint matrix based on the driving model, identifies a dominant air mass path, combines pollutant concentration data of receptor points, constructs and verifies a random forest regression model, takes the footprint matrix as features, takes standardized concentration as labels, quantifies the contribution of grid points through feature importance and Shapley explanation methods, and identifies key pollution source areas. The application can achieve the technical effect of significantly improving the accuracy and efficiency of atmospheric pollution tracing and providing more scientific and accurate decision-making basis for atmospheric pollution prevention and control.

Description

Technical Field

[0001] This application relates to the field of atmospheric pollution source tracing technology, and in particular to an atmospheric pollution source tracing method and system that integrates numerical models and artificial intelligence algorithms. Background Technology

[0002] As a core component of refined management and control, the accuracy and efficiency of source tracing directly determine the targeting and effectiveness of pollution control measures. Currently, complex regions such as urban clusters and industrial parks face the combined influence of multiple factors, including diverse pollution source types, scattered spatial distribution, and dynamic fluctuations in emission intensity, significantly increasing the difficulty of source tracing. This also places higher demands on the anti-interference capabilities, interpretability of results, and quantitative accuracy of source tracing technologies.

[0003] Current mainstream air pollution source tracing technologies can be divided into three main categories, each with significant shortcomings: First, traditional statistical source tracing methods rely on correlation analysis between monitoring data and emission inventories. While simple to operate, they are prone to misclassifying external environmental impacts as contributions from pollution sources, resulting in low accuracy and failing to meet the precise requirements of actual supervision. Second, AI-based predictive source tracing methods use deep learning and machine learning models to fit the patterns of pollutant concentration changes, improving accuracy compared to traditional methods. However, the "black box" nature of these models means the results lack clear physical mechanism support, making it difficult for environmental regulators to formulate targeted control measures. Third, numerical model simulation source tracing methods combine meteorological fields and diffusion models to analyze pollution transmission paths. While supported by physical mechanisms, they lack adaptability to dynamically changing pollution sources and struggle to quantify the causal contribution relationship between individual pollution sources and receptor sites.

[0004] Furthermore, existing source tracing technologies mostly focus on single aspects such as prediction and source tracing, making it difficult to adapt to real-world source tracing scenarios involving complex regional pollution source dynamics and multi-factor coupling interference. Based on the cross-integration of artificial intelligence and numerical models, this technology overcomes the "correlation trap" and "black box" bottlenecks of traditional technologies, constructing a high-precision, interpretable, and highly interference-resistant integrated source tracing technology. This not only directly addresses the core pain points of current precise pollution control but also provides scientific and technological support for environmental supervision and enforcement, industrial layout optimization, and emission reduction plan formulation, possessing significant industry application value and practical guiding significance. Summary of the Invention

[0005] The purpose of this application is to provide an atmospheric pollution source tracing method and system that integrates numerical models and artificial intelligence algorithms, which can achieve accurate source tracing of atmospheric pollutants at receiver points and identification of key source areas with spatial interpretability.

[0006] To achieve the above objectives, this application provides the following solution.

[0007] In the first aspect, this application provides a method for tracing the source of air pollution that integrates numerical models and artificial intelligence algorithms, including the following steps.

[0008] The meteorological data are preprocessed and high-precision meteorological field numerical simulations are performed. The high-precision meteorological field simulation data is then converted into a data format that can be read by the stochastic time inversion Lagrange diffusion model.

[0009] Based on the transformed high-precision meteorological field data, the random time inversion Lagrange transport model is driven to generate a high-resolution spatiotemporal footprint matrix and identify the dominant air mass transport path.

[0010] After standardizing the atmospheric pollutant concentration data of the receptor points with the high-resolution spatiotemporal footprint matrix, a random forest regression model is constructed and the model performance is verified. The high-resolution gridded spatiotemporal footprint matrix is ​​used as the feature, and the standardized pollutant concentration of the receptor points is used as the output label. The receptor points are air quality monitoring stations or environmentally sensitive points within the target area.

[0011] By quantifying the contribution of each grid point in the high-resolution gridded spatiotemporal footprint matrix to the prediction results of the random forest regression model using feature importance analysis and Shapley additive interpretation, key source regions that have a significant impact on the concentration of atmospheric pollutants at the receptor point are identified.

[0012] Secondly, this application provides an atmospheric pollution source tracing system that integrates numerical models and artificial intelligence algorithms, including the following modules.

[0013] The high-precision meteorological field numerical simulation module is used to preprocess meteorological-related basic data and perform high-precision meteorological field numerical simulations. It converts the high-precision meteorological field simulation data obtained from the simulation into a data format that can be read by the stochastic time inversion Lagrange diffusion model.

[0014] The dominant air mass transport path module is used to drive the stochastic time inversion Lagrange transport model to generate a high-resolution spatiotemporal footprint matrix based on the transformed high-precision meteorological field data, and to identify the dominant air mass transport path.

[0015] The random forest regression model training module is used to construct a random forest regression model and perform model performance verification after standardizing the atmospheric pollutant observation concentration data of the receptor point and the high-resolution spatiotemporal footprint matrix. The high-resolution gridded spatiotemporal footprint matrix is ​​used as the feature, and the standardized pollutant concentration of the receptor point is used as the output label. The receptor point is an air quality monitoring station or environmentally sensitive point in the target area.

[0016] The key source region identification module is used to quantify the contribution of each grid point in the high-resolution gridded spatiotemporal footprint matrix to the prediction results of the random forest regression model through feature importance analysis and Shapley additive interpretation method, and to identify key source regions that have an important impact on the concentration of atmospheric pollutants at the receptor point.

[0017] According to the specific embodiments provided in this application, this application has the following technical effects.

[0018] This application provides a method and system for tracing atmospheric pollution sources by integrating numerical models and artificial intelligence algorithms. By combining numerical models and artificial intelligence algorithms, this method can fully utilize meteorological reanalysis data, underlying surface information, and observational data to achieve high-precision meteorological field simulation, thereby generating a high-resolution spatiotemporal footprint matrix. The application of cluster analysis technology makes the identification of the dominant air mass transport path more accurate, providing strong support for subsequent pollution source tracing. Simultaneously, the random forest regression model constructed using the footprint matrix as a feature can effectively fit pollutant concentrations, and the accuracy of the model is ensured through validation. Finally, by quantifying the contribution of each grid point to the prediction results, key source areas can be accurately identified, providing targeted measures for atmospheric pollution control. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an application environment diagram of an air pollution source tracing method that integrates numerical models and artificial intelligence algorithms according to an embodiment of this application.

[0021] Figure 2 This is a flowchart illustrating an air pollution source tracing method that integrates numerical models and artificial intelligence algorithms, provided as an embodiment of this application.

[0022] Figure 3 This is a schematic diagram of the functional modules of an air pollution source tracing system that integrates numerical models and artificial intelligence algorithms, provided as another embodiment of this application.

[0023] Figure 4 The following is an overall flowchart of an air pollution source tracing method that integrates numerical models and artificial intelligence algorithms, provided as another embodiment of this application.

[0024] Figure 5 The distribution of the high-resolution spatiotemporal footprint matrix simulation results provided in step S22 of another embodiment of this application.

[0025] Figure 6 Step S23, provided in another embodiment of this application, shows the clustering results of the dominant air mass transport path.

[0026] Figure 7 Step S32, provided in another embodiment of this application, verifies the performance of the random forest regression model.

[0027] Figure 8 Step S41, provided in another embodiment of this application, preliminarily locks the feature importance parameters and key source regions based on the random forest model.

[0028] Figure 9 Step S42, provided in another embodiment of this application, generates a contribution map of key influence source regions based on SHAP values.

[0029] Figure 10 Step S43, provided in another embodiment of this application, is a spatial statistical summary diagram based on the distribution of SHAP values.

[0030] Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0032] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] The air pollution source tracing method integrating numerical models and artificial intelligence algorithms provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on other servers. Terminal 102 can send the meteorological-related basic data to be processed to server 104. After receiving the meteorological-related basic data, server 104 preprocesses it and performs high-precision meteorological field numerical simulation, converting the simulated high-precision meteorological field data into a data format readable by the stochastic time inversion Lagrange diffusion model. Based on the converted high-precision meteorological field data, the stochastic time inversion Lagrange transport model is driven to generate a high-resolution spatiotemporal footprint matrix and identify the dominant air mass transport path. After standardizing the receiver point atmospheric pollutant observation concentration data and the high-resolution spatiotemporal footprint matrix, a stochastic forest regression model is constructed and its performance is verified; the high-resolution gridded spatiotemporal footprint matrix is ​​used as the feature, and the standardized receiver point pollutant concentration is used as the output label. The contribution of each grid point in the high-resolution gridded spatiotemporal footprint matrix to the prediction results of the random forest regression model is quantified by feature importance analysis and Shapley additive interpretation methods, identifying key source regions that have a significant impact on the concentration of air pollutants at the recipient points. Server 104 can then feed back the identified key source regions to terminal 102. Furthermore, in some embodiments, the air pollution source tracing method integrating numerical models and artificial intelligence algorithms can also be implemented independently by server 104 or terminal 102. For example, terminal 102 can directly process the meteorological-related basic data to be processed, or server 104 can obtain the meteorological-related basic data to be processed from the data storage system and process it accordingly.

[0034] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0035] In one exemplary embodiment, such as Figure 2 As shown, a method for tracing the source of air pollution integrating numerical models and artificial intelligence algorithms is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1Taking server 104 as an example, the explanation includes the following steps A1 to A4. Wherein: Step A1 involves preprocessing the meteorological-related basic data and performing high-precision meteorological field numerical simulations. The high-precision meteorological field simulation data obtained is then converted into a data format that can be read by the stochastic time inversion Lagrange diffusion model.

[0036] Step A2: Based on the transformed high-precision meteorological field data, drive the stochastic time inversion Lagrange transport model to generate a high-resolution spatiotemporal footprint matrix and identify the dominant air mass transport path.

[0037] Step A3: After standardizing the atmospheric pollutant concentration data of the receptor point and the high-resolution spatiotemporal footprint matrix, a random forest regression model is constructed and the model performance is verified. The high-resolution gridded spatiotemporal footprint matrix is ​​used as the feature, and the standardized pollutant concentration of the receptor point is used as the output label. The receptor point is an air quality monitoring station or an environmentally sensitive point within the target area.

[0038] Step A4 involves quantifying the contribution of each grid point in the high-resolution gridded spatiotemporal footprint matrix to the prediction results of the random forest regression model through feature importance analysis and Shapley additive interpretation, thereby identifying key source regions that have a significant impact on the concentration of atmospheric pollutants at the receptor points.

[0039] Implementing steps A1 to A4 above quantifies the contribution of grid points to model predictions, identifies key source areas, and provides a basis for pollution control, enabling precise source tracing of air pollutants. Furthermore, this application can further improve the accuracy and efficiency of source tracing by continuously optimizing the fusion method of the numerical model and artificial intelligence algorithms. Simultaneously, optimizing the parameter settings and feature selection methods of the random forest regression model enhances the model's fitting effect on changes in pollutant concentrations. Through these optimization measures, the air pollution source tracing method integrating numerical models and artificial intelligence algorithms provided in this application will be better able to adapt to the source tracing needs of different regions and different pollution types, providing more precise technical support for air pollution control.

[0040] In another exemplary embodiment of this application, in order to perform high-precision meteorological field numerical simulation and obtain a data format readable by the random time inversion Lagrange diffusion model, the above step A1 is replaced by the following steps A11 to A16.

[0041] Step A11: Obtain basic meteorological data; the basic meteorological data includes global meteorological reanalysis data, underlying surface data, and observational data; the observational data is hourly atmospheric pollutant concentration data at the receptor point. In this embodiment, the global meteorological reanalysis data specifically refers to NCEP-FNL global meteorological reanalysis data, with a horizontal resolution of 1.0°×1.0° and a temporal resolution of 6 hours, including core meteorological elements such as wind, temperature, humidity, pressure, and mixing layer height; the underlying surface data specifically includes static data such as topography, land use type, and surface albedo. Observational data: hourly atmospheric pollutant observation data from air quality monitoring stations.

[0042] Step A12 involves preprocessing the underlying surface data using a data preprocessing system to obtain the geographic grid data file. In this embodiment, step A12 specifically involves using the geogrid module in the WPS data preprocessing system to set aspects of the model grid, such as projection, simulation range, resolution, and nesting relationships, and interpolating the terrain and other underlying surface data onto the model grid to obtain the geographic grid data file.

[0043] Step A13 involves preprocessing the global meteorological reanalysis data and observational data using a data preprocessing system to obtain the meteorological grid data file. Specifically, the ungrb module in the WPS system is used to extract and decode the required meteorological variables from the global meteorological reanalysis data. The metgrid module in the WPS system is then used to perform spatiotemporal interpolation on the decoded meteorological variables and match them to the model grid set by geogrid to obtain the meteorological grid data file.

[0044] Step A14 involves batch initializing the geographic grid data file and the meteorological grid data file to obtain an initial field file and a boundary field file. In this embodiment, the batch initialization process specifically involves using the data initialization program real to batch initialize the files.

[0045] Step A15 involves performing a high-precision meteorological field simulation using the meteorological grid data file, initial field file, and boundary field file to obtain high-precision meteorological field simulation data. The core elements of this high-precision meteorological field simulation data include hourly U / V wind components, temperature, air pressure, mixing layer height, and boundary layer height. In this embodiment, a mesoscale meteorological model (WRF) is used for the high-precision meteorological field simulation.

[0046] Step A16: The high-precision meteorological field simulation data obtained from the simulation is converted into a data format readable by the Stochastic Time Inversion Lagrange Diffusion Model (STILT) to obtain binary grid data. In this embodiment, a dedicated data interface conversion program, wrf2stilt, is used to convert the high-precision meteorological field simulation data output by the WRF model into binary grid data readable by the STILT stochastic time inversion Lagrange diffusion model.

[0047] In another exemplary embodiment of this application, in order to identify the dominant air mass transport path by using a random time inversion Lagrange transport model, step A2 above is replaced by steps A21 to A23.

[0048] Step A21 involves setting key parameters for the stochastic time inversion Lagrange transport model. In this embodiment, step A21 specifically includes: determining a receptor point within the target area and setting particle release; setting the receptor point height to a first predetermined height; setting the particle release frequency to a first predetermined frequency and the reverse tracking duration to a first predetermined duration; and setting the vertical height of particle diffusion to a second predetermined height. In this embodiment, the receptor point can be single or multiple; the first predetermined height is 2 meters; the first predetermined frequency is 1000 particles / hour; the first predetermined duration is 72 hours; and the second predetermined height is 3000 meters, covering the main atmospheric layers within the boundary layer.

[0049] Step A22: Based on the binary grid data, drive the random time inversion Lagrange transfer model to generate a high-resolution gridded spatiotemporal footprint matrix.

[0050] Step A23: Perform cluster analysis on the high-resolution gridded spatiotemporal footprint matrix, and use the silhouette coefficient as the criterion for determining the optimal number of clusters to obtain the dominant air mass transport path.

[0051] In another exemplary embodiment of this application, step A22 is replaced by steps B1 to B4.

[0052] Step B1: Initialize the simulation region, spatial grid, and simulation time length, and define the core calculation rules for particle backtracking and footprint matrix simulation to obtain the model initialization results.

[0053] Step B2: Based on the model initialization results, load the binary grid data to simulate the hourly spatiotemporal positions of each particle released from the receptor point during the atmospheric backtracking period within a first set duration. In this embodiment, the first set duration is 72 hours.

[0054] Step B3: By statistically analyzing the key gridding parameters of the stochastic time inversion Lagrange transport model, the statistical results of the key gridding parameters are obtained; the key gridding parameters specifically include: effective number of particles, grid dwell time and spatial distribution frequency.

[0055] Step B4: Generate a high-resolution gridded spatiotemporal footprint matrix based on the statistical results of key gridding parameters.

[0056] In another exemplary embodiment of this application, step A23 is replaced by steps C1 to C2.

[0057] Step C1: The K-means clustering algorithm is used to perform cluster analysis on the high-resolution gridded spatiotemporal footprint matrix to obtain the silhouette coefficients.

[0058] Step C2 involves using the contour coefficient as the criterion for determining the optimal number of clusters, and setting the optimal number of clusters K as the first predetermined number of categories based on atmospheric transport characteristics, thereby obtaining the dominant air mass transport path of the first predetermined number of categories. In this embodiment, the first predetermined number of categories is specifically 3 to 5.

[0059] In another exemplary embodiment of this application, in order to construct a random forest regression model and complete the model performance verification, the above step A3 is replaced by the following steps A31 to A35.

[0060] Step A31 involves standardizing the observed atmospheric pollutant concentration data at the receiver site with the high-resolution spatiotemporal footprint matrix to obtain a standardized high-resolution spatiotemporal footprint matrix and standardized observed atmospheric pollutant concentrations at the receiver site. In this embodiment, the standardization process eliminates dimensional differences.

[0061] Step A32 involves precisely aligning the standardized observation data with the standardized high-resolution spatiotemporal footprint matrix in terms of time dimension to obtain a time-aligned dataset. In this embodiment, the precise alignment in the time dimension ensures that each time-series atmospheric pollutant concentration observation corresponds to a unique footprint matrix.

[0062] Step A33: Divide the time-aligned dataset into a training set and a test set according to a first predetermined ratio. In this embodiment, the first predetermined ratio is specifically 7:3.

[0063] Step A34: Using the standardized high-resolution gridded spatiotemporal footprint matrix as feature values ​​and the standardized atmospheric pollutant observation concentration at the receptor point as the output label, a random forest regression model is constructed and trained on the training set to obtain the trained random forest regression model. In this embodiment, the specific parameter settings of the random forest regression model are as follows: the number of decision trees is set to 200, the maximum depth is set to None, the minimum number of sample splits is set to 2, and the minimum number of sample leaf nodes is set to 1.

[0064] Step A35: Apply the trained random forest regression model to the test set and use multi-dimensional core validation metrics to quantify the fitting accuracy and prediction reliability of the trained random forest regression model.

[0065] In another exemplary embodiment of this application, step A35 is replaced by steps D1 to D2.

[0066] Step D1: Apply the trained random forest regression model to the test set, and use the correlation coefficient, root mean square error, and standardized mean deviation as core validation metrics to obtain the core validation metric quantification model.

[0067] Step D2: Validate the overall performance of the trained random forest regression model using the core validation metrics quantification model.

[0068] In another exemplary embodiment of this application, in order to perform high-precision meteorological field numerical simulation and obtain a data format readable by the random time inversion Lagrange diffusion model, the above step A4 is replaced by the following steps A41 to A45.

[0069] Step A41: Evaluate the influence of each grid point in the standardized high-resolution gridded spatiotemporal footprint matrix on the prediction results of the random forest regression model based on the feature importance parameter, and determine the contribution.

[0070] Step A42: Based on the contribution, the key source regions that have a significant impact on the concentration of atmospheric pollutants at the receiver point are initially identified spatially.

[0071] Step A43: Using the Shapley additive interpretation method, calculate the SHAP value of each atmospheric pollutant concentration prediction sample corresponding to the standardized high-resolution gridded spatiotemporal footprint matrix; the positive or negative sign of the SHAP value represents the direction of contribution of the corresponding grid to the atmospheric pollutant concentration at the receiver point, and the absolute value reflects the degree of contribution.

[0072] Step A44 involves spatially visualizing the SHAP value of each predicted atmospheric pollutant concentration sample to generate a high-resolution source region contribution map. In this application, the contribution map can transform the "black box" predictions of the machine learning model into a spatially interpretable contribution distribution.

[0073] Step A45: Select grids whose absolute SHAP value is greater than a first set contribution threshold as key source regions. In this embodiment, the first set contribution threshold is specifically the value at the 80th percentile after all grids' absolute SHAP values ​​are sorted from largest to smallest.

[0074] Based on the same inventive concept, this application also provides an air pollution source tracing system for implementing the aforementioned fusion of numerical models and artificial intelligence algorithms. The solution provided by this system is similar to the implementation scheme described in the above method; therefore, the specific limitations of one or more embodiments of the air pollution source tracing system fusion of numerical models and artificial intelligence algorithms provided below can be found in the limitations of the air pollution source tracing method fusion of numerical models and artificial intelligence algorithms described above.

[0075] In one exemplary embodiment, such as Figure 3 As shown, an air pollution source tracing system integrating numerical models and artificial intelligence algorithms is provided, including: The high-precision meteorological field numerical simulation module is used to preprocess meteorological-related basic data and perform high-precision meteorological field numerical simulations. It converts the high-precision meteorological field simulation data obtained from the simulation into a data format that can be read by the stochastic time inversion Lagrange diffusion model.

[0076] The dominant air mass transport path module is used to drive the stochastic time inversion Lagrange transport model to generate a high-resolution spatiotemporal footprint matrix based on the transformed high-precision meteorological field data, and to identify the dominant air mass transport path.

[0077] The random forest regression model training module is used to construct a random forest regression model and perform model performance verification after standardizing the atmospheric pollutant observation concentration data of the receptor point and the high-resolution spatiotemporal footprint matrix. The high-resolution gridded spatiotemporal footprint matrix is ​​used as the feature, and the standardized pollutant concentration of the receptor point is used as the output label. The receptor point is an air quality monitoring station or environmentally sensitive point in the target area.

[0078] The key source region identification module is used to quantify the contribution of each grid point in the high-resolution gridded spatiotemporal footprint matrix to the prediction results of the random forest regression model through feature importance analysis and Shapley additive interpretation method, and to identify key source regions that have an important impact on the concentration of atmospheric pollutants at the receptor point.

[0079] As an optional implementation method, an air pollution source tracing method integrating numerical models and artificial intelligence algorithms is also provided, wherein implementing this implementation method, Figure 4This paper illustrates the complete technical solution of the source tracing method from data input to source area output, which includes four core steps: "meteorological data preprocessing and high-precision meteorological field numerical simulation" (S1), "high-resolution spatiotemporal footprint matrix simulation" (S2), "random forest regression model construction and verification" (S3), and "key impact source area identification" (S4). The data flow between each step is marked with arrows to clarify the connection between the steps. At the same time, the key sub-links of each step are marked, intuitively presenting the technical logic closed loop covering "numerical simulation - transmission process analysis - key source area identification", which facilitates the understanding of the overall implementation process of the invention. Each step is elaborated in detail below.

[0080] S1. Meteorological data preprocessing and high-precision numerical simulation of meteorological fields.

[0081] The core of this step is to simulate high-precision meteorological field data for the target area. The specific steps are as follows.

[0082] S11. Basic data collection and preprocessing.

[0083] Global meteorological reanalysis data: NCEP-FNL global meteorological reanalysis data, with a horizontal resolution of 1.0°×1.0° and a temporal resolution of 6 hours, including core meteorological elements such as wind, temperature, humidity, pressure, and mixed layer height.

[0084] Underlying surface data: static data such as topography, land use type, and surface albedo. Observational data: hourly atmospheric pollutant observation data from air quality monitoring stations.

[0085] Data preprocessing operations: The geogrid module in the WPS data preprocessing system is used to set model grid settings such as projection, simulation range, resolution, and nesting relationships. Topographic and other underlying surface data are interpolated onto the model grid to obtain a geographic grid data file. The ungrib module in the WPS system is used to extract and decode the required meteorological variables from global meteorological reanalysis data. The metgrid module in the WPS system performs spatiotemporal interpolation on the decoded meteorological variables and matches them to the model grid set in geogrid to obtain a meteorological grid data file.

[0086] S12, High-precision meteorological field simulation.

[0087] Data initialization: The data initialization program real is used to perform batch initialization processing on the obtained geographic grid data files and meteorological grid data files to generate initial field files and boundary field files.

[0088] Numerical simulation: The mesoscale meteorological model WRF is used, and the obtained meteorological grid data file, initial field file and boundary field file are input to perform high-precision meteorological field simulation.

[0089] High-precision meteorological field simulation data output: The simulation outputs the core elements of a high-precision meteorological field, including incrementally smaller U / V wind components, temperature, air pressure, mixing layer height, boundary layer height, etc.

[0090] S13, Meteorological data format conversion.

[0091] The high-precision meteorological field simulation data output by the WRF model is converted into binary grid data that can be read by the stochastic time inversion Lagrange diffusion model STILT using the dedicated data interface conversion program wrf2stilt.

[0092] S2, High-resolution spatiotemporal footprint matrix simulation.

[0093] The core of this step is to drive the Stochastic Time Inversion Lagrange Transmission (STILT) model to generate a high-resolution spatiotemporal footprint matrix and identify the dominant air mass transport path. The specific steps are as follows.

[0094] S21, STILT model key parameter settings.

[0095] Receptor point determination: Select air quality monitoring stations or environmentally sensitive points within the target area as receptor points; a single receptor point or several receptor points can be set; the height of the receptor point is set to 2m.

[0096] Particle release settings: Particles are released from the receiver at a frequency of 1000 particles / hour, and the reverse tracking time is set to 72 hours to fully characterize the randomness of atmospheric motion; the vertical height of particle diffusion is set to 3000 meters to cover the main atmospheric layers within the boundary layer.

[0097] S22, Simulation of the Spatiotemporal Footprint Matrix.

[0098] Model initialization: Load the binary grid data that the STILT model can read after S13 processing, and initialize the simulation region, spatial grid and simulation time length, etc., and define the core calculation rules for particle backtracking and footprint matrix simulation.

[0099] STILT model simulation: Based on the model initialization results, the simulation begins with the hourly spatiotemporal positions of each particle released from the receptor point during a 72-hour atmospheric backtracking period. Simultaneously, the STILT model statistically analyzes key gridding parameters such as the effective number of particles, grid dwell time, and spatial distribution frequency, and generates a high-resolution gridded spatiotemporal footprint matrix based on the statistical results of these key gridding parameters.

[0100] S23, Identification of the dominant air mass transport path.

[0101] K-means clustering algorithm was used to perform clustering analysis on high-resolution gridded spatiotemporal footprint matrix. The silhouette coefficient was used as the criterion for determining the optimal number of clusters. Combined with atmospheric transport characteristics, the optimal number of clusters K was set to 3~5, and 3~5 dominant air mass transport paths were identified accordingly.

[0102] S3. Construction and validation of the random forest regression model.

[0103] The core of this step is to construct a random forest regression model using a high-resolution gridded spatiotemporal footprint matrix as eigenvalues ​​to fit the observed concentrations of atmospheric pollutants at the receptor points, and then validate the random forest regression model. The specific steps are as follows.

[0104] S31. Data time alignment and preprocessing.

[0105] Time alignment: The hourly atmospheric pollutant concentration data of the receptor point are precisely aligned with the high-resolution spatiotemporal footprint matrix generated by S22 in the time dimension to ensure that the atmospheric pollutant concentration data of each time point corresponds to a unique footprint matrix.

[0106] Dataset partitioning: The time-aligned dataset is divided into a training set and a test set in a 7:3 ratio. The training set is used for model training, and the test set is used for model performance verification.

[0107] Standardization processing: The atmospheric pollutant observation concentration data and the high-resolution gridded spatiotemporal footprint matrix are standardized to eliminate dimensional differences.

[0108] S32. Random Forest Regression Model Training and Quantitative Validation.

[0109] Model parameter settings: number of decision trees set to 200, maximum depth set to None, minimum number of sample splits set to 2, minimum number of sample leaf nodes set to 1.

[0110] Model training process: The high-resolution gridded spatiotemporal footprint matrix after standardization is used as the feature value, and the observed concentration of atmospheric pollutants after standardization of the receptor point is used as the output label. The random forest regression model is trained on the training set to achieve fitting and prediction of atmospheric pollutant concentration at the receptor point.

[0111] Model performance validation: The trained model is applied to the test set to validate its performance. The correlation coefficient (R), root mean square error (RMSE), and standardized mean deviation (NMB) are used as core validation metrics. Each metric is used to quantify the model’s fitting accuracy and prediction reliability of atmospheric pollutant concentrations at the receptor point, thus completing the overall performance validation of the random forest regression model.

[0112] S4. Identification of key impact source areas.

[0113] The core of this step is to quantify the contribution of the footprint values ​​of each grid point in the spatiotemporal footprint matrix to the pollutant concentration prediction results of the random forest regression model using the Shapley additive interpretation method (SHAP), thereby identifying key source regions that have a significant impact on the atmospheric pollutant concentration at the receiver point. The specific steps are as follows.

[0114] S41. Key impact source areas have been preliminarily identified.

[0115] During the training of the S32 random forest regression model, the output feature importance parameter is automatically calculated. This parameter is used to evaluate the influence of each input feature value (i.e., the standardized high-resolution gridded spatiotemporal footprint matrix) on the prediction results of the random forest regression model (i.e., the standardized predicted or fitted values ​​of pollutant concentrations at the receptor site). The higher the feature importance parameter value, the greater the contribution of that feature value to the prediction results. Based on this, key source regions that have a significant impact on the atmospheric pollutant concentrations at the receptor site can be initially identified spatially.

[0116] S42. Generation of contribution maps of key impact source regions.

[0117] SHAP value calculation: The Shapley additive interpretation method (SHAP) is used to calculate the SHAP value of each atmospheric pollutant concentration prediction sample corresponding to the high-resolution gridded spatiotemporal footprint matrix (i.e. model input feature value) after standardization. The positive or negative sign of the SHAP value represents the direction of the contribution of the corresponding grid to the atmospheric pollutant concentration of the receiver point (positive value indicates increased concentration, negative value indicates decreased concentration), and the absolute value reflects the degree of contribution.

[0118] Contribution map generation: Spatially visualize the SHAP value of each grid to generate a high-resolution source contribution map, transforming the "black box" prediction of the machine learning model into a spatially interpretable contribution distribution.

[0119] S43, Spatial Statistical Key Impact Source Areas.

[0120] Contribution threshold setting: Select the top 80% of grids by absolute SHAP value as the key influence source area.

[0121] Matching key impact source regions with dominant transport paths: Combine the dominant air mass transport paths identified by S23 to compare whether key impact source regions are located on the dominant transport paths.

[0122] Quantitative results output: Based on the contribution map of the S42 key source area, a quantitative analysis at the administrative unit scale is completed through spatial statistical methods, and the contribution of each region (province, autonomous region, municipality) to the concentration of air pollutants at the receiver point is output.

[0123] Figure 5Centered on the receptor point, the average footprint matrix of the virtual particle over a 72-hour simulation period is shown. The transmission sensitivity of each grid is represented by the color intensity (the darker the color, the stronger the transmission sensitivity), corresponding to the footprint matrix generation step S22 (the unit of the filled part is: ). Figure 6 This is a visualization of the K-means clustering analysis of the high-resolution spatiotemporal footprint matrix. Based on the geographical base map of the study area, wind fields corresponding to the clusters are overlaid with different colors to label the transport distribution of four dominant air masses. Each cluster is numbered and its corresponding transport probability is given, clearly showing the main transport direction of air masses from the upstream source region to the receiver point. This corresponds to the clustering analysis results of S23 (the units of the footprint clustering color-coded part are: ). Figure 7 A scatter plot comparing the model fitting results, with the horizontal axis representing the ozone concentration observed at the receptor site (μg / m³). 3 The vertical axis represents the concentration predicted by the random forest model (μg / m³). 3 The figure also marks the specific values ​​of the fitted line, correlation coefficient (R), root mean square error (RMSE), mean absolute error (NMB), and sample size (N). Through the distribution relationship between observed and predicted values, it can be clearly shown that the model can accurately capture ozone concentration changes based on footprints, corresponding to the model validation stage in S32. Figure 8 Includes two related subplots: the left side shows observed ozone concentrations greater than 0 μg / m³. 3 In the case of (all samples), the spatial distribution of feature importance parameters is based on the target region grid, and the magnitude of influence is characterized by color gradient (the darker the color, the greater the influence, but the direction of contribution cannot be determined); the right side shows the observed ozone concentration greater than 100 μg / m³. 3 Under these circumstances, the spatial distribution of feature importance parameters is analyzed. The key influence source region corresponding to S41 is preliminarily identified. Figure 9 Includes two related subplots: the left side shows observed ozone concentrations greater than 0 μg / m³. 3 (For all samples), the spatial distribution of SHAP values ​​(source contribution map) is based on a grid of the study area, and the direction and degree of contribution are represented by red and blue two-color gradients (red represents positive contribution, blue represents negative contribution, and the darker the color, the greater the contribution); the right side shows the observed ozone concentration greater than 100 μg / m³. 3 Under these conditions, a spatial distribution map of SHAP values ​​(source contribution map) is generated. This corresponds to the generation of the contribution map of the key influence source region in S42. Figure 10 Spatial statistical methods were used to present a quantitative analysis at the administrative unit scale, outputting the contribution of each key region (province, autonomous region, municipality) to the concentration of air pollutants at the receiver site. This clearly identifies the regions with the greatest impact on ozone concentration prediction.

[0124] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 11 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data on key source areas. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it can implement the atmospheric pollution source tracing method integrating numerical models and artificial intelligence algorithms provided in the previous embodiment.

[0125] Those skilled in the art will understand that Figure 11 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0126] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0127] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0128] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0129] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0130] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0131] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0132] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for tracing the sources of air pollution that integrates numerical models and artificial intelligence algorithms, characterized in that, include: Preprocessing of meteorological-related basic data and high-precision numerical simulation of meteorological fields are performed. The high-precision meteorological field simulation data obtained is then converted into a data format that can be read by the stochastic time inversion Lagrange diffusion model. Based on the transformed high-precision meteorological field data, the stochastic time inversion Lagrange transport model is driven to generate a high-resolution spatiotemporal footprint matrix and identify the dominant air mass transport path. After standardizing the atmospheric pollutant concentration data of the receptor points with the high-resolution spatiotemporal footprint matrix, a random forest regression model is constructed and the model performance is verified. The high-resolution gridded spatiotemporal footprint matrix is ​​used as the feature, and the standardized pollutant concentration of the receptor points is used as the output label. The receptor points are air quality monitoring stations or environmentally sensitive points within the target area. By quantifying the contribution of each grid point in the high-resolution gridded spatiotemporal footprint matrix to the prediction results of the random forest regression model using feature importance analysis and Shapley additive interpretation, key source regions that have a significant impact on the concentration of atmospheric pollutants at the receptor point are identified.

2. The method for tracing the source of air pollution by integrating numerical models and artificial intelligence algorithms according to claim 1, characterized in that, Preprocessing of meteorological-related basic data and performing high-precision numerical simulations of meteorological fields are carried out. The simulated high-precision meteorological field data is then converted into a data format readable by the stochastic time inversion Lagrange diffusion model, specifically including: Acquire basic meteorological data; the basic meteorological data includes global meteorological reanalysis data, underlying surface data, and observational data; the observational data is hourly atmospheric pollutant concentration data at the receptor point. The underlying surface data is preprocessed using a data preprocessing system to obtain the geographic grid data file; The global meteorological reanalysis data and observation data are preprocessed using a data preprocessing system to obtain the meteorological grid data file; The geographic grid data file and the meteorological grid data file are batch initialized to obtain the initial field file and the boundary field file; High-precision meteorological field simulation is performed using the meteorological grid data file, initial field file, and boundary field file to obtain high-precision meteorological field simulation data. The core elements of the high-precision meteorological field simulation data include: hourly U / V wind components, temperature, air pressure, mixing layer height, and boundary layer height. The high-precision meteorological field simulation data obtained from the simulation is converted into a data format readable by the stochastic time inversion Lagrange diffusion model to obtain binary grid data.

3. The method for tracing the source of air pollution by integrating numerical models and artificial intelligence algorithms according to claim 1, characterized in that, Based on the transformed high-precision meteorological field data, a stochastic time inversion Lagrange transport model is driven to generate a high-resolution spatiotemporal footprint matrix and identify the dominant air mass transport path, specifically including: Set the key parameters for the stochastic time inversion Lagrange transport model; Based on the binary grid data, a random time inversion Lagrange transfer model is driven to generate a high-resolution gridded spatiotemporal footprint matrix; Cluster analysis was performed on the high-resolution gridded spatiotemporal footprint matrix, and the silhouette coefficient was used as the criterion for determining the optimal number of clusters to obtain the dominant air mass transport path.

4. The method for tracing the source of air pollution by integrating numerical models and artificial intelligence algorithms according to claim 3, characterized in that, Setting key parameters for the stochastic-time inversion Lagrange transport model specifically includes: Determine the receptor point within the target area and set the particle release; set the receptor point height to a first preset height; set the particle release frequency to a first preset frequency and the reverse tracking duration to a first preset duration; set the vertical height of particle diffusion to a second preset height.

5. The method for tracing the source of air pollution by integrating numerical models and artificial intelligence algorithms according to claim 3, characterized in that, Based on the binary grid data-driven stochastic time inversion Lagrange transfer model, a high-resolution gridded spatiotemporal footprint matrix is ​​generated, specifically including: Initialize the simulation region, spatial grid, and simulation time length, and define the core calculation rules for particle backtracking and footprint matrix simulation to obtain the model initialization results; Based on the model initialization results, the binary grid data is loaded to simulate the hourly spatiotemporal position of each particle released from the receptor point during the atmospheric backtracking period in the first set time period; By statistically retrieving the key gridding parameters of the stochastic time inversion Lagrange transport model, the statistical results of the key gridding parameters are obtained; the key gridding parameters specifically include: effective particle number, grid dwell time, and spatial distribution frequency; Based on the statistical results of key gridding parameters, a high-resolution gridded spatiotemporal footprint matrix is ​​generated.

6. The method for tracing the source of air pollution by integrating numerical models and artificial intelligence algorithms according to claim 3, characterized in that, Cluster analysis is performed on the high-resolution gridded spatiotemporal footprint matrix, using the silhouette coefficient as the criterion for determining the optimal number of clusters, to obtain the dominant air mass transport path, specifically including: The K-means clustering algorithm was used to perform cluster analysis on the high-resolution gridded spatiotemporal footprint matrix to obtain the silhouette coefficients; The contour coefficient is used as the criterion for determining the optimal number of clusters, and the optimal number of clusters K is set as the first set number of categories based on atmospheric transport characteristics, thereby obtaining the dominant air mass transport path of the first set number of categories.

7. The method for tracing the source of air pollution by integrating numerical models and artificial intelligence algorithms according to claim 1, characterized in that, After standardizing the atmospheric pollutant concentration data at the receptor point with the high-resolution spatiotemporal footprint matrix, a random forest regression model was constructed and its performance was validated. Specifically, this included: After standardizing the atmospheric pollutant observation concentration data at the receptor point and the high-resolution spatiotemporal footprint matrix, we obtain the standardized high-resolution spatiotemporal footprint matrix and the standardized atmospheric pollutant observation concentration at the receptor point. The standardized observation data is precisely aligned with the standardized high-resolution spatiotemporal footprint matrix in terms of time dimension to obtain a time-aligned dataset. The time-aligned dataset is divided into a training set and a test set according to a first predetermined ratio; Using the standardized high-resolution gridded spatiotemporal footprint matrix as the feature value and the standardized atmospheric pollutant observation concentration at the receptor point as the output label, a random forest regression model is constructed and trained on the training set to obtain the trained random forest regression model. The trained random forest regression model was applied to the test set, and multi-dimensional core validation metrics were used to quantify the fitting accuracy and prediction reliability of the trained random forest regression model.

8. The method for tracing the source of air pollution by integrating numerical models and artificial intelligence algorithms according to claim 7, characterized in that, The trained random forest regression model was applied to the test set, and multi-dimensional core validation metrics were used to quantify the fitting accuracy and prediction reliability of the trained random forest regression model, specifically including: The trained random forest regression model was applied to the test set, and the correlation coefficient, root mean square error, and standardized mean deviation were used as core validation metrics to obtain a quantitative model of core validation metrics. The overall performance of the trained random forest regression model is verified by quantifying the model using the core validation metrics.

9. The method for tracing the source of air pollution by integrating numerical models and artificial intelligence algorithms according to claim 1, characterized in that, The contribution of each grid point in the high-resolution gridded spatiotemporal footprint matrix to the prediction results of the random forest regression model was quantified using feature importance analysis and Shapley additive interpretation methods. Key source regions with significant influence on atmospheric pollutant concentrations at the receptor sites were identified, specifically including: The influence of each grid point in the standardized high-resolution gridded spatiotemporal footprint matrix on the prediction results of the random forest regression model is evaluated based on the feature importance parameters, and the contribution is determined. Based on the aforementioned contribution, key source regions that have a significant impact on the concentration of atmospheric pollutants at the receiver site are initially identified spatially; The Shapley additive interpretation method is used to calculate the SHAP value of each atmospheric pollutant concentration prediction sample corresponding to the high-resolution gridded spatiotemporal footprint matrix after standardization. The positive or negative sign of the SHAP value represents the direction of contribution of the corresponding grid to the atmospheric pollutant concentration at the receiver point, and the absolute value reflects the degree of contribution. Spatial visualization of the SHAP value of each atmospheric pollutant concentration prediction sample is performed to generate a high-resolution source region contribution map. Grids with an absolute SHAP value greater than the first set contribution threshold are selected as key source regions.

10. An air pollution source tracing system integrating numerical models and artificial intelligence algorithms, characterized in that, The air pollution source tracing system that integrates numerical models and artificial intelligence algorithms includes: The high-precision meteorological field numerical simulation module is used to preprocess meteorological-related basic data and perform high-precision meteorological field numerical simulation, converting the high-precision meteorological field simulation data obtained into a data format that can be read by the stochastic time inversion Lagrange diffusion model. The dominant air mass transport path module is used to drive the stochastic time inversion Lagrange transport model to generate a high-resolution spatiotemporal footprint matrix based on the transformed high-precision meteorological field data, and to identify the dominant air mass transport path. The random forest regression model training module is used to construct a random forest regression model and perform model performance verification after standardizing the atmospheric pollutant observation concentration data of the receptor point and the high-resolution spatiotemporal footprint matrix. The high-resolution gridded spatiotemporal footprint matrix is ​​used as the feature, and the standardized pollutant concentration of the receptor point is used as the output label. The receptor point is an air quality monitoring station or environmentally sensitive point in the target area. The key source region identification module is used to quantify the contribution of each grid point in the high-resolution gridded spatiotemporal footprint matrix to the prediction results of the random forest regression model through feature importance analysis and Shapley additive interpretation method, and to identify key source regions that have an important impact on the concentration of atmospheric pollutants at the receptor point.

Images